Data Analytics in Manufacturing Funding Sources : NIST, DMG Mori Seiki, System Insights Raunak Bhinge © 2014 LMAS contact email: [email protected] Objectives ■ LMAS has identified ‘Big Data’ applications in manufacturing as an important research area in the field of ‘Sustainable Manufacturing.’ Introduction ■ With the increase in data availability and sensor fusion in the manufacturing industry, the role of data analytics in the development of smart manufacturing systems has grown tremendously. ■ The main objectives of the on-going research in this field are: ■ ■ ■ ■ ■ ■ ■ ■ Effective machine tool characterization Real-time process monitoring Intelligent parameter selection Energy-efficient tool path planning Improvement of machine tool efficiency Sustainable manufacturing techniques Intelligent tool life estimation and tool selection Applications of data analysis in long term machine tool selection, planning and maintenance Experimental Setup ■ 9 test parts were machined using a full factorial DOE, which involved a wide range of processes like face milling, contouring, pocketing, slotting, spiraling and drilling. ■ Manufacturing enterprises are striving towards greener and more energy-efficient manufacturing systems with the use of smart machine tools. ■ The current project deals with the prediction of energy consumption in machine tools as a means of effective machine tool characterization in order to develop more energy-efficient tool paths and cutting strategies without compromising tool life. Data Extraction Agent ■ The data extraction agent post-processes the data extracted from the MTConnect Agent by simulating the cutting process. ■ Block-wise data is generated after cutting simulation, data condensation and transformation. ■ Power consumption and machining data was collected using MTConnect, followed by tool wear measurements. Energy Prediction Using the Gaussian Process ■ The Gaussian Process, which employs the Gaussian Mixture Model, is used to predict the energy consumption Feature Analysis ■ Effects of individual features like machining direction and milling strategy are analyzed using the predicted energy function, in order to identify strategies for energy reduction. ■ The relative total error in prediction was observed to be 1.34% for the global GP model and 1.93% for the collective GP model Real-time Data Processing & Energy Prediction ■ An architecture for real-time data processing and an adaptive energy prediction model is being developed and implemented. Future Work ■ Develop the real-time data processing and implement the adaptive energy prediction algorithm. ■ Identify important energy-impacting features in order to determine new energy-efficient cutting strategies. ■ Investigate the effect of tool wear and prediction of the same using graphical modeling techniques with latent variables. ■ Develop an API for CAM software which use the developed models to predict tool wear and energy consumption for a chosen set of process parameters, cutting strategy, tool path strategy and workpiece orientation. Identify an optimized tool paths and process parameters for an energy-efficient process. Berkeley / UNIVERSITY OF CALIFORNIA